The "95% Win Rate EA" Red Flag, Explained

A 95% win rate is usually a warning sign, not proof an EA works. High win rates are mostly produced by martingale, averaging down, or tiny targets with huge stops, so a few big losses wipe out many small wins. Ask for the average loss and the open drawdown.

Browse any EA marketplace and you'll find EAs advertising win rates of 90%, 95%, even higher. To most buyers this reads as quality: surely an EA that wins 19 trades out of 20 must work? In automated trading the logic runs the other way. An extreme win rate is one of the most reliable warning signs on any sales page, and the reasons are pure arithmetic.

Win rate says nothing without the other half

Expectancy - what a strategy actually earns per trade - is win rate multiplied by average win, minus loss rate multiplied by average loss. A 95% win rate with small winners is wiped out by rare large losers: win 95 trades at £10 and lose 5 at £200 and you're down. A 40% win rate with winners three times the size of losers is comfortably positive. Win rate alone is a numerator missing its denominator - quoting it in isolation is the tell.

How EAs manufacture high win rates

• Martingale and averaging-down. The classic route. By adding to losing positions until a bounce closes the basket positive, almost every basket "wins" - while the losses accumulate invisibly until one trend realises them all at once. The full maths is in why martingale EAs eventually fail, and grid systems use the same disguise.
• Tiny targets, giant stops. Take profit at 5 pips, stop-loss at 200, and most trades win - mechanically. Each rare stop-out erases dozens of winners. The win rate is real; it's just measuring the wrong thing.
• Curve-fitting. Tune enough parameters against history and you can produce almost any statistic, win rate included. It describes the past by construction - overfitting and walk-forward testing explained covers why it evaporates on new data.
• Cherry-picked windows and small samples. Fifty trades over a favourable quarter can show any win rate at all. Sample size is the difference between a statistic and an anecdote.

The questions that puncture the number

1. What are the average win and average loss, in pips and money?
2. What's the worst single loss and the maximum drawdown - including open drawdown?
3. Does position size ever increase after a loss?
4. Over how many trades, and were any out-of-sample?
5. Does the live record - if one exists - show the same profile as the backtest? It usually doesn't.

An honest vendor answers all five without flinching. A win-rate merchant changes the subject.

Build instead of buy: our biased take

Declared bias: we make Nebula, a no-code builder that breeds strategies with a genetic algorithm and validates them walk-forward, out-of-sample and blind forward. Relevant here: when you build and validate a strategy yourself, you see the whole statistical picture - expectancy, loss distribution, drawdown - not one flattering number chosen for a sales page. Some honestly-validated strategies will have modest win rates; that's what real ones often look like. We promise no particular numbers at all, which is rather the point.

The bottom line

A 95% win rate isn't evidence an EA works; it's usually evidence of how it's hiding its risk. Ask for the average loss and the open drawdown, and watch the magic trick collapse. The rest of the toolkit is in evaluating an EA before buying.

Trading carries a high level of risk and automated systems do not eliminate it. Nothing here is financial advice; past performance does not predict future results.